CCS Students Shine on the Global Stage with Innovative Research

In recent years, students from the College of Computer Science at National Yang Ming Chiao Tung University (NYCU) have made remarkable achievements across diverse fields including artificial intelligence, computer architecture, and knowledge systems. Their research has been presented at top-tier international conferences, showcasing exceptional academic strength and creativity.
A particularly noteworthy milestone this year is Prof. Tsung-Tai Yeh's team publishing a paper at the International Symposium on Computer Architecture (ISCA) — one of the most prestigious conferences in the field and a rare accomplishment for NYCU in nearly two decades. This success highlights the College's depth in high-performance computing and systems design research.
From deep learning and image generation to semantic ontology and knowledge structures, as well as high-efficiency computing architectures, NYCU students demonstrate interdisciplinary innovation and global engagement. Their participation in world-leading venues such as ICLR, ISCA, and FOIS not only reflects their dedication and capability but also underscores the College's commitment to cultivating the next generation of outstanding researchers.

Title: Reinvigorating Structured Knowledge and Ontologies for Trustworthy and Beneficial AI and Robotics
Authors: Chun-yien Chang, Yi-Ting Chen, and Ying-ping Chen
Advisor: Prof. Ying-ping Chen
Conference: The International Conference on Formal Ontology in Information Systems (FOIS)

Significance:
The Formal Ontology in Information Systems (FOIS) conference is internationally recognized as the leading venue for foundational research in ontology. It plays a central role in advancing theoretical and methodological work on upper ontology, a field that underpins semantic interoperability, explainable AI, and the design of trustworthy knowledge infrastructures. FOIS brings together experts across disciplines—computer science, philosophy, information systems, and artificial intelligence—establishing a rare forum where conceptual rigor and practical system design converge. Its long-standing reputation and impact make it a cornerstone event for shaping how ontological foundations inform the next generation of intelligent and interoperable systems.

The experience of Chun-yien Chang:
In Taiwan, the application of ontology in research is not rare, yet the lack of foundational work and the near absence of engagement with the international community have left this field with little visibility, even as grounded semantics and ontology-based interoperability are becoming indispensable in the AI era. Against this backdrop, the acceptance of our position paper into the main track of FOIS represents a rare milestone for a Taiwanese institution. The paper was conceived not merely as a presentation, but as a call from our local vantage point: to urge that, beyond advancing individual projects, we must collectively recognize ontology's potential contributions at the policy level, especially as AI and robotics begin to reshape human life on an unprecedented scale. The responses and encouragement I received from senior researchers and practitioners around the world after the conference affirmed both the timeliness of this perspective and the possibility of building meaningful connections. This experience has encouraged me to continue developing this difficult but vital field, with the hope that this work may contribute to its growth first and foremost within the domestic environment—despite the challenges that remain.

Title: Ranking-aware Adapter for Text-driven Image Ordering with CLIP
Authors: Wei-Hsiang Yu, Yen-Yu Lin, Ming-Hsuan Yang, and Yi-Hsuan Tsai
Advisor: Prof. Yen-Yu Lin
Conference: International Conference on Learning Representations (ICLR)

Significance:
The International Conference on Learning Representations (ICLR) is one of the world's leading conferences in artificial intelligence, recognized alongside NeurIPS and ICML as a premier venue for groundbreaking research. It focuses on deep learning, representation learning, and related fields, with an innovative open review process that fosters transparency and collaboration. Attracting top researchers from academia and industry leaders such as Google DeepMind, OpenAI, and Meta AI, ICLR is a hub for presenting state-of-the-art methods that often shape the future of machine learning. Acceptance here signifies high academic impact, innovation, and global recognition in the AI research community.

The experience of Wei-Hsiang Yu:
I am deeply grateful to Prof. Lin and all collaborators for their guidance throughout the research, writing, and review process. In this work, we proposed a novel approach that integrates textual input with images and designs a contrastive mechanism to enable multi-image, multi-task ranking. Having our paper accepted by ICLR was an important milestone in my academic journey.
During the conference, I had the chance to discuss and exchange ideas with researchers from around the world, gaining new insights and discovering potential interdisciplinary applications. I was particularly impressed by how dedicated many attendees were — conducting experiments and writing papers even during the sessions — which reminded me of the value of persistence and focus in research, and inspired me to work even harder toward excellence.

 

Title: Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation
Authors: Sheng-Feng Yu, Jia-Jiun Yao, and Wei-Chen Chiu
Advisor: Prof. Wei-Chen Chiu
Conference: The Thirteenth International Conference on Learning Representations (ICLR 2025)

Significance:
The International Conference on Learning Representations (ICLR) is one of the world's leading conferences in artificial intelligence, recognized alongside NeurIPS and ICML as a premier venue for groundbreaking research. It focuses on deep learning, representation learning, and related fields, with an innovative open review process that fosters transparency and collaboration. Attracting top researchers from academia and industry leaders such as Google DeepMind, OpenAI, and Meta AI, ICLR is a hub for presenting state-of-the-art methods that often shape the future of machine learning. Acceptance here signifies high academic impact, innovation, and global recognition in the AI research community.

The experience of Sheng-Feng Yu:
Attending ICLR 2025 was an unforgettable experience. It felt surreal to be surrounded by so many brilliant minds in machine learning, each sharing cutting-edge ideas that are shaping the future of the field. I am truly thankful to Prof. Chiu for his mentorship and encouragement—without his support, I wouldn't have had the chance to bring our work to such an incredible platform.
What I enjoyed the most were the conversations outside the presentations. Talking with other researchers gave me new perspectives on our work, and I was surprised by how many people showed genuine interest in our method. Those exchanges not only validated the effort we've put in but also sparked fresh ideas I hadn't considered before.
Beyond the academic side, ICLR also reminded me why I love doing research—it's not just about solving problems, but about being part of a community that pushes each other forward. The experience left me both humbled and motivated, and I returned home with a stronger drive to keep exploring.

Title: Efficient Action-Constrained Reinforcement Learning via Acceptance-Rejection Method and Augmented MDPs

Authors: Wei Hung, Shao-Hua Sun, Ping-Chun Hsieh

Advisor: Prof. Ping-Chun Hsieh

Conference: The 13th International Conference on Learning Representations (ICLR 2025)

Significance:

The International Conference on Learning Representations (ICLR) is one of the world's premier conferences in machine learning, recognized alongside NeurIPS and ICML as a top venue for cutting-edge research in deep learning, representation learning, and artificial intelligence. The 2025 edition marked a special milestone as ICLR returned to Asia (Singapore) for the first time in years, drawing leading researchers, data scientists, and industry experts from North America, Europe, and Asia. Being accepted for presentation at ICLR signifies exceptional innovation and impact at both algorithmic and theoretical levels, offering an invaluable opportunity to engage directly with the global AI research frontier.

 

The experience of Wei Hung:

I was deeply honored to attend ICLR 2025 in Singapore, the first in-person conference held in Asia in recent years. It was an extraordinary journey that allowed me to fully immerse myself in a world-class academic atmosphere. During the poster session, I had the opportunity to present our research on a general and efficient framework for Action-Constrained Reinforcement Learning (ACRL), designed to achieve zero action violations in safety-critical applications such as robotic control. More importantly, I had the privilege of engaging in profound discussions with global experts, gaining valuable inspiration throughout the event.

 

Professional reflection and deep exchange:

I had in-depth conversations with researchers from top international institutions such as Google and IBM, focusing on topics like multi-objective decision learning and experimental design in the context of action constraints. These face-to-face discussions went beyond technical exchange—they encouraged me to reflect critically on potential blind spots in my own research and opened new perspectives for leveraging multi-objective learning architectures to improve the robustness of single-objective training.

 

Emerging trends and cross-domain insights:

This year's conference was heavily dominated by topics surrounding Large Language Models (LLMs). I participated in several RL×LLM sessions and observed many pioneering approaches that integrate LLMs into the decision-making loop as policy planners or reward models. These insights provided valuable guidance for my future research on integrating LLMs into reinforcement learning, and they deepened my understanding that cross-disciplinary collaboration has become an inevitable direction in modern AI research.

 

Technology translation and research vision:

Through this journey, I realized that the value of academic research lies not only in publishing papers but also in continuously innovating and translating technologies into real-world impact. This experience encouraged me to broaden my research perspective toward more complex and practical scenarios, striving to enhance both the performance and applicability of algorithms in real-world contexts.

 

I am sincerely grateful to my teammates for their hard work and to our project sponsors for their support. This experience not only allowed our research to be showcased on an international stage but also infused my academic journey with renewed energy and purpose.

Title: AQB8: Energy-Efficient Ray Tracing Accelerator through Multi-Level Quantization
Authors: Yen-Chieh Huang, Chen-Pin Yang, and Tsung-Tai Yeh
Advisor: Prof. Tsung-Tai Yeh
Conference: International Symposium on Computer Architecture (ISCA)

Significance:
The International Symposium on Computer Architecture (ISCA) is one of the most prestigious conferences in computer architecture, featuring the latest and most innovative research on processor design, memory systems, accelerators, and parallel computing architectures. The 2025 acceptance rate was 22%.

The experience of Chen-Pin Yang:
I am deeply grateful to Prof. Yeh and my teammate Yen-Chieh for their tireless collaboration, which made it possible for us to represent NYCU at ISCA — one of the world's premier computer architecture conferences. In our paper, we proposed a novel ray tracing acceleration structure that provides an effective solution to current challenges in computation and performance efficiency. Presenting our work to leading industry players such as AMD, Intel, and Huawei, and engaging with top researchers worldwide, was both inspiring and enlightening. This experience broadened my horizons and sparked new ideas for future designs, further strengthening my motivation to pursue innovation in this field.